Improving Query by Tapping via Tempo Alignment

Query by tapping (QBT) is a content trieval method that can retrieve a song by taking the u er’s tapping or clapping at the note onsets of the intended song in the database for comparison. This paper proposes a new query-by-tapping algorithm that aligns the IOI (i ter-onset interval) vector of the query sequence with songs in the dataset by build ing an IOI ratio matrix, and then applies a dynamic programming (DP) method to compute the optimum path with minimum cost. Exper ments on different datasets indicate that our algorithm outperforms other previous approaches in accuracy 10 and MRR), with a speedup factor of 3 in computation. With the advent of personal handheld devices, QBT pr vides an interesting and innovative way for music retrie al by shaking or tapping the devices, which is also di cussed in the paper.